◊ SYME = / ‘YES’
/ ‘NOT’ [DEFECT]
This parameter is common to all the linear solveurs (direct, iterative). It is appealable only for the nonlinear operators.
Warning : The translation process used on this website is a "Machine Translation". It may be imprecise and inaccurate in whole or in part and is
The interest of this keyword is to save time at the time of the resolution of the linear systems because the direct solveurs are generally more expensive in nonsymmetrical.
The symmetrization of the matrix
K
conduit however to solve a different linear system. By doing this, it can degrade the convergence of the non-linear algorithms.◊ NPREC = / nprec
/ 8 [DEFECT]
◊ STOP_SINGULIER = / ‘YES’ [DEFECT]
/ ‘NOT’
These two parameters are common to all the direct linear solveurs (LDLT, MULT_FRONT, MUMPS).
They are used to control the course of digital factorization and the quality of the solution of the linear system. The digital factorization of a matrix can fail in two cases: problem of construction of factorized (structurally or numerically singular matrix) and digital detection of a singularity (solution of the unstable linear system).
The keyword NPREC and STOP_SINGULIER allow to fix the threshold of detection of the singularities and the behavior to be adopted in the event of failure during factorization.
nprec is used to gauge the process of detection as singularity of the matrix of the system to be solved. With LDLT and MULT_FRONT, one takes the absolute value of nprec, with MUMPS, one takes nprec because its sign has an importance: if
nprec <0
, the detection of singularity is disabled, if not it is activated.In all the cases, if the value nprec to zero one is left initializes it with the value by default (8).
By initializing this parameter has a rather low value (1 or 2) (respectively strong, for example, 20), the detection of singularity will very often start (respectively seldom).
For LDLT , MULT_FRONT :
When at the end of factorization, one notes that a diagonal term
d '
became very small (compared to what it was before factorizationd
), it is that the matrix is (probably) almost singular. That is to sayn=log ∣ d ' d ∣, this report magnitude indicates that on an equation (at least) one lost n
significant
figures.
If
n> nprec
, one idiotsidère that the matrix is singular. If the user indicated:STOP_SINGULIER=' OUI' The code stops then in error.
STOP_SINGULIER='NOT‘ The execution continues after emission of one ALARM. The quality of the solution is then not guaranteed. This parameter setting is not advised. Into nonlinear, it is not inevitably too prejudicial with the quality of the results because those “are corrected” by the process of Newton.
For MUMPS :
So for at least a pivot, the infinite standard of the line (or column) is lower than the threshold
10
−nprecthen the matrix is regarded as singular.
Warning : The translation process used on this website is a "Machine Translation". It may be imprecise and inaccurate in whole or in part and is provided as a convenience.
One compares some aspects of the two types of criteria of detection of singularity in documentation [U2.08.03].
Note:
• Any important loss of significant figures during a factorization is an indicator of a badly posed problem. Several causes are possible (nonexhaustive list): boundary conditions insufficient of blocking of the structure, linear relations redundant, very heterogeneous numerical data (too large terms of penalization)…
• For LDLT and MULT_FRONT, the detection of singularity is made all the time because it is far from expensive.
• Concerning MUMPS, a mechanism makes it possible to check the quality of the solution in addition (RESI_RELA). One thus left freedom disable this criterion (by choosing one nprec negative).
• By default, with the direct solvor MUMPS, one thus has a double quality control of the solution: into linear, RESI_RELA and NPREC, into nonlinear, the criterion of Newton and NPREC. It is possible to disconnect them, but it is not advised without good reason.
◊ ELIM_LAGR = ‘NOT’/‘YES’/‘LAGR2’
This keyword makes it possible to eliminate the equations from Lagrange corresponding to the conditions dualized kinematics.
By default (except for MUMPS), these equations are not eliminated (‘NOT‘).
Technique of elimination used for ELIM_LAGR= 'YES‘is described in [R3.03.05]. Attention, this technique does not function in parallel.
In the case of the solvor MUMPS, a third value is possible: ‘LAGR2‘. The objective is then to remove the second equation of Lagrange, but to preserve the first.
The value ‘LAGR2‘is the defect for solvor MUMPS.
Historically, direct linear solveurs of Code_Aster (‘MULT_FRONT' and ‘LDLT’) did not have algorithm of swivelling. To circumvent this problem, the taking into account of the limiting conditions by of Lagranges was modified by introducing doubles Lagranges at the cost of a overcost report and calculation. Like MUMPS have faculties of swivelling, this choice of dualisation of the limiting conditions can be called into question.
This parameter can be temporarily disabled by the code not to distort the calculation of the determinant of the matrix. This functionality is mainly necessary by the operators CALC_MODES with OPTION among [‘NEAR’, ‘ADJUSTS’, ‘SEPARATE’] and INFO_MODE. The user is then informed of this automatic change of parameter setting via a message dedicated (only in INFO=2).
Warning : The translation process used on this website is a "Machine Translation". It may be imprecise and inaccurate in whole or in part and is